Hi, everyone. My name is Simon Langevin, VP of product at Kopeo, and welcome to this webinar about the evolution of the search box in commerce and the future of search in retail and ecommerce. Now the first question you're asking yourself would be, you know, what even drives this evolution? What are we talking about? And, obviously, the quick answer would be generative AI. You know, AI comes to mind, but, really, generative AI is the thing that is currently evolving, the way that we interact with, brands, with retail websites, and and with the search. So here's the example of, you know, our traditional Amazon search bar and their new Rufus, the the GIF here on the left, the new kind of chatbot shopping assistant solution that they've now released to the American audience. But before, you know, we go into the the disturbance or any sorts of disruption caused by things such as Rufus, let's step back and look at, you know, what is generative AI. And and, obviously, you know, it's been there for a few years, so we know the basic of it. But I wanna mostly expand on the application of Gen AI when it comes to discovery. So, obviously, the, you know, the main description here, generative AI is a large language model. So it is a specific type of machine learning models which use language generation or or use its knowledge of human language in order to do generation. So it creates text in a form of generative AI, but can also do, you know, what we've seen, audio, video, even produce games and whatnot. But in the case of discovery, we're mostly interested around the generation of text and mostly the understanding of text as well. So this idea of being able to feed it any sorts of any sorts of human language and then being responded by that. The top player out there, obviously, ChatGPT from OpenAI. We also know about Copilot from Google, Gemini from from Google. Sorry. Copilot by Microsoft, Perplexity, and Clode, which is by Intropic. There's there's tons of others, but these are the one that we see the most as general GenAI. So mostly, you know, that you can use in in multiple use cases in order to produce text to help you review code or or create emails or or whatnot. Perplexity itself, however, I started to offer something very specific related to retail or or to shopping, which is a AI powered shopping assistant. And, obviously, the question is, you know, is this pretty much the end of, you know, shopper that will go to a website and then, you know, search for a product and then purchase that product? So is it is it the end of that typical kind of conversion journey that we're used to? And the answer so far is not really. It's actually more of a new channel that will compete with what we call the first search for product. And and right now, that first search for product is is really kind of a a fight between, I would say, three main players, Google, Amazon. Obviously, Amazon is actually the winner of that first search for now. It's around in the US, last time I reviewed, it was around fifty four percent of the first query for product. When someone looks for a product, the first query that they make is typically on Amazon. But more you know, what we see growing more and more is obviously social media. And in some countries such as in China, for example, TikTok will be more than a social media hacks media app. It will be a super app where you also do your ecommerce and your shopping. So we see kind of these three channel which will lead you to the conversion path, which will lead you to whether, you know, a website which can finish the conversion, or in the case of Amazon, you'll have the a to c, pretty much the entire conversion, and even the fulfillment happens by Amazon. So Amazon is obviously it's kind of own little silo, but Google, social media, and at least in in North America, And now these kind of new, you know, Perplexity, ChatGPT will all fall into that kind of ecosystem of first discovery, which then brings you to a retailer, a brand that will finish the transaction. So it opens the door, however, for, you know, these brands once you have the the shopper that reached their their their site. What can they do with GenAI? Because perplexity, even if, you know, I'm being replaced myself as a shopper, I'm being replaced by a robot and assistant that will do shopping for me. You know, that's pretty much the vision here. To be at, like, at the time, you know, in the early two thousand ten when we had Alexa, talking about, you know, just shopping for you on, after after you tell it, like, the the voice call. But now it's it's, you know, that robot that will do, the shopping for me. But there's still there's still need to be someone to serve that robot. Right? Who's gonna wait on the brand website, you know, to serve to service that that robot or to respond to social media traffic and and Google traffic? The brands. Right? You know, apart again from that kind of Amazon closed silo, everyone else is kind of dependent on social media, Google, or in this case, that new channel that is GenAI chat solution. So what have been, you know, some of the tests being done? So these are these were really early. Some of the first failure that we saw with with GenAI, Chevrolet, a local dealership, which, you know, tried a kind of ChatGPT powered chatbot with absolutely no grounding whatsoever, and it just completely went rogue. Started to sell, you know, full on cars for one dollar or from ten dollars, and, you know, just kinda saying, yeah. Sure. You can buy, you know, Chevrolet or even saying that, you know, some competing brand were better. Like, you know, oh, yes. For an electric car or, you know, Tesla has more experience or something like this. Obviously, not something that Chevrolet wants, you know, the the the the the public to hear about, and it's not entirely true as well. You know? I mean, we it's it's all debatable. Air Canada in in Canada went, also had issues, and it went actually a bit beyond. It actually ended up in lawsuit where, you know, you had some customers that were told that, you know, they could take a certain trip or, you know, that they had some some options attached to their trips, which were actually not true. So this ended up in lawsuits, which are actually being solved now. You know? It's it's still, in the news, although this happened almost, I think, almost three years ago. Now more recently and related more to retail, we have Rufus. So Amazon, you know, I said it's in a closed silo, but even Amazon themselves, you know, they invest into GenAI for their own solution within Amazon. And, again, here, you know, some early failure that we saw. You know, someone is asking about Coca Cola, and it asked, you know, what what are my options to buy Coca Cola? And mostly the answer is, well, maybe you should take Pepsi, because, you know, Pepsi is is better for you. It's actually healthier for you, which, again, extremely debatable. I don't think anyone thinks that Pepsi is in any way healthier than Coca Cola or vice versa. But, you know, this was one of the, what we call, hallucination of the the GenAI solution. It leads me to the adoption of RAG, retrieval augmented generation. And this is where, you know, I'm kind of circling back now to the search experience. So what is RAG? RAG is retrieval augmented generation. It's an AI framework, so it lives around the GenAI model or the LLM, and it proves the quality of the response by grounding it into external source of of knowledge as well as filters, security trimming, and and everything pretty much that the search does today. So when you use a modern search engine, there is a respect of the permission you have access to, the filters that, for example, merchandise we want to put in place, the boost on certain product, the brewery on certain other product. And this is all respected by the large language model when RAG is being used. So a search engine, which usually, you know, just the feature of a typical search engine on a website would be, obviously, search and suggestion will drive your listing and your recommendation, but on top of this, can be used for retrieval augmented generation. So the the the same technology that power the search will be used, to make sure that you ground your Gen AI experience. Some of these technology, just to to give you an example here, this is kind of the model of layered AI by by Caveil. So they're very specific to Caveil, but it's a model that is used, you know, by by multiple different vendor in in in the in in similar or different ways. So you have, you know, a precision element to it, which is, you know, a lexical match. So I'm searching for a blue shirt. I expect the word blue and shirt to be considered or, in this case, a gray hoodie. There is a notion of semantics. So I understand that gray, you know, is a color. And even though it might not be in the actual title of, you know, the the the product itself, I can find it within the description or the image or just even by similar keywords. So blue and teal, for example, would be things that, you know, could be considered somewhat similar. Yeah. Behavioral learning, which will use simply clickstream. So a product is click often and leads to a purchase. Well, it's is probably more popular than another one. And vector search, a more recent approach, which, vector search is used for for multiple element. But in the case of Kaveo, we use it both for semantic but also for intent detection. So to be able to detect that, people who are looking at a certain subset of product, are interested into that other set of product because they are somewhat similar or complementary in what we call the vector space. So vector search itself will would warrant a a full on webinar on it, but it is mostly a way for the search engine to recreate the catalog in the way that it understands it. So instead of having a catalog, for example, that is sorted by categories, it is sorted by intent. So people that usually look for, you know, a gray hoodie might be looking for sweatpants. So kind of, you know, common intent that we can look for. So with all of these technologies in place, a a search engine like Aveo can do what we call RAD. So, you know, the first stage retrieval is simply rules that you can manually put. So security trimming, again, if you want to make sure that you don't that you don't show product that people don't don't have access to. Also, filtering, just the fact that you will start recommending products that are part of your catalog. And this is quite important in the case of the Chevrolet dealership, for example. It started to recommend other cars from other manufacturer. This is something that would not happen with Rag. You also have your boost and brewery merchandising rule, which I must not forget about, but mostly, you know, your generative experience will actually respect the boost that you have put in your main search. Then you have hybrid relevance, which is everything I've just talked about, so lexical match, semantics, vector search intent detection. And then from there, you know, the products are returned by the search engine just like a normal search that you would do on Amazon or on any other website, but the search engine will retrieve the chunks. It will retrieve mostly the most important elements of these products. You know, it will understand in the description that, for example, the products that you're looking for are shirts or or sweaters. It will understand that, you know, these sweaters comes in different colors or a different shape, different sizes. So these are all important elements. And these chunks are processed by the search engine, and then they are fed to the large language model. The large language model can be, you know, the the GPT three point five or four point zero, could be Cohere, could be Microsoft Azure, GPT, which is pretty much the same model as OpenAI. So it could be all of these different models, which are then, you know, fed by the search engine. And then it will return and and then the search engine will also add additional instructions for prompt. This is where it will say, for example, do not forget that you are a Chevrolet dealership, hence the idea of the brand name Chevrolet is extremely important. And once all of this is done, then the large language model returns what we call a grounded response. So a good example of that here, on the left side is Dell dot com. So Dell, which sells laptop, electronic components, and, you know, computers and even video games, console, mostly video game console. And on the right side is ChatGPT four point o. And I'm asking the same question on both sides. How can I get better at gaming? You know? How can I stop losing online, against my friends? And, you know, on the right side, you can see that ChatGPT takes more of a general answer, ungrounded answer. Not bad, actually. No hallucination there. It's a very kind of safe question. But it says you mostly, you know, set some goals, develop your understanding of the game, you know, the mechanics and the strategy the the strategies of your game. While on the left side, it is grounded in the fact that Dell is a manufacturer, distributor, a retailer of electronics and components and laptop and whatnot. So it start to talk about hardware requirement, graphic driver. You should update your graphic driver. And, by the way, we sell this. You know, you can consider a hardware upgrade. And, know, we sell as SSD. You could be interested in it. Use GPU optimization tool, which we sell, and etcetera, etcetera. So you can see here both answers are pretty good. Actually, no dangerous hallucination on this side. But on that side, it's actually, it it's actually tied, you know, to what I want. And if I and the other interesting is here you can see, you know, high level streamer for for pro player. Here, customize your key binding, mouse sensibility, and and controller. You know, if it started to recommend product here, it will probably go beyond Dell, which is obviously not what Dell will want. So using just ChatGPT as is for Dell is extremely dangerous, hence why grounding. But another important thing is because it's now controlled by the underlying search engine, the facets, the sorts, everything that you expect from a search solution does also impact the retrieval. Meaning that in this case, what I did is I simply used the facet for the knowledge base instead of, you know, the the much more larger dataset of manual and document. And, automatically, you know, the retrieval was changed and started to, you know, include elements such as, you know, regular practice, learning from others, staying updated. So all of this is quite interesting because it shows that, you know, as a as a shopper, I I I'm grounded, but, also, you know, I can still use the typical element of a search engine to get where I want. So it is another big advantage of of Rag. And here, I gave an example where, you know, it was inside of the search at Dell. However, Rag can be used pretty much in every other system, such as, for example, a chatbot, you know, instead of instead of the search or recommendation, could be in a chatbot. And this leads to the idea of the chatbot. Because if we go back to, you know, the original intent of that webinar, the title, it started really with the evolution of the search, and then, you know, on one side, Rufus, the other side, you have the Amazon search. So which one is better? You know, should we should we go to a chatbot or not? Is chatbot even relevant? So let's ask ChatGPT. Right? So the the the I asked the same question that I asked at the start. What is generative AI, you know, in two sentences or less? And this is what ChatGPT responded. And funny enough, you can see that even ChatGPT has biased toward the idea of chatbot. So response to prompt enabling application like chatbot image generation and content creation. But you can see that chatbots was there from the start. So it does consider itself a chatbot and does consider that GenAI is a chatbot thing. And this is quite interesting because there's no GenAI is is much more than chatbot. Actually, it does chat text generation, and you can see that the first definition that I've put at the start was not from ChatGPT. It was just from Wikipedia and and even for actually a combination of of Wikipedia and other scientific articles. But in this case here, it is from the chatbot itself. So there is a bias which has been created mostly by the fact that it learns from all of us, from our own biases as human, and we have the bias of chat. And why? Because we like robots. And it's a bit it's a bit of a, you know, kind of a subjective way to to to talk about it. But when we think about AI, we always think about robots. We think about that idea that we will have human like individual that would, at one point, you know, converse with us and and and live among us in a way, you know, and and even with the idea that they will have a human shape, you know, and they'll be robots. So we expect that kind of conversational experience, you know, with this chat bot. We we kind of already visualize ourselves discussing with that AI, with with the the the, yeah, that concept of a conversational AI. And and, automatically, you know, if we take our existing interfaces, it leads us to, you know, thinking about that chatbot experience. And and more and more now, you know, on the market, because chatbot were already a thing that existed, didn't necessarily have the best of reputation. You know, buying using a chatbot was never such a great thing. It's kinda been rebranded as assistant or shopping assistant. You know? So that idea of, you know, being assisted, instead of chatting. You know? But it's still it's still pretty much a chatbot, you know, which has maybe a better technology underneath it. So this has led, obviously, to an explosion of what we call, you know, the best AI shopping assistant, which has pretty much led now, you know, back to that original conversation of we will no longer search. We will use assistant. Right? We will, you know, be assisted. We will be in a kind of conversation and whatnot. So there has been, you know, tons of them, tons of new start up that have released shopping assistant. There have been existing player who have released them. But if we look just at this article here, you know, from from Shop Dev, there were tons of brands that I personally even never heard about. So new startups that have just started so, definitely, there's a trend here. There's a a potential disruption happening on the market when there's so so much money being put, you know, for for new entrants on the market. But let's let's look at the big guy. So, you know, before we jump into the small player, like, what what is the big the big one doing? You know, Amazon Rufus, is it working or not? According to Amazon, it is. According to Amazon, you know, they they suggest seven hundred million financial gain from its AI shopping assistant, Rufus. And, obviously, you know, I I don't know all of the financials of Amazon, but it's not a negligible amount. Right? But then when you read the actual article, there's some interesting elements to It generates no direct revenue, which is kind of expected. You know, it's not it's not a product that they sell. It's actually a mean to an end, so I'm fine with that. But it says that, you know, it includes mostly the income from ad place within Rufus' response to inquiries. So most of the money so far that they expect to receive from that is not so much on the increase in conversion, but mostly from the advertisement that they'll get out of this product. And this is quite interesting because if I am any other brand who doesn't have, you know, the retail media power of Amazon, then, you know, it's not necessarily a good sign. Right? It's it doesn't necessarily mean that it will be worth my investment. And on a usability standpoint, you know, I gave kind of the the the funny example of Coca Cola versus Pepsi at the start, but, you know, let's look a little bit larger at a micro level. Is it working or not? Well, Amazon is not giving the actual numbers when it comes to conversion usage or or or anything related to that beyond the hype, you know, of the first few days. They don't give these numbers, but there has been tons and tons and tons and tons of reviews. And so far, both for sellers as well as for buyers, it's not been received well. Actually, it's been received as pretty much not personalized, not understanding, not really converting, So not a the right place necessarily to invest in. And and, you know, there's an element of technology behind that. Like, you know, is it just Gen AI is not good? And I do not believe it's the case. Because if it works, you know, for Perplexity, for ChatGPT, why wouldn't it work for for Amazon? I believe that definitely the technology underneath it, it's good. The problem is not the technology. It's the interface. And, you know, instead of going into all of the, you know, articles about the technology itself or or or anything like that, I decided to go for something around the UX, so the user experience. This is from N and G. They offer training and and all that. I don't know them. I just like to read them because, usually, what they write about is extremely relevant. And they went from a a very naive we don't know the technology. We are just gonna look at it from a user standpoint. We're gonna do a review of Amazon Rufus. And it led to a fairly interesting part where they say, you know, valuable yet invisible. And they said that it was, you know, hard to find Rufus in a way where, you know, it's not in our typical mindset to think about shopping into the small box on the left that is the chat. Right? We're we're more interested into the discovery, being able to see where the product are, so being able to search or navigate through the categories. But even once you actually found it, what they said is that it has a tendency to get confused about what the user is shopping for. Technology can be solved. Give unnecessarily worthy and jargon filled response, and that part is very important and brings me back to the idea of the robot. We want pretty much everyone that has demoed anything related to ChatGPT inside of shopping, inside of retail. You know, if you have been attending any other webinar about, you know, GenAI in shopping or you went to events about, you know, the the new things around GenAI, pretty much everything has been repeated. You start as a user. You start a conversation with that AI, and there's a lot of back and forth before you get to any actual products, and before you get to enough discovery so that you're confident with your choice. So, again, here, our mental mindset of putting forward a robot or or robot like experience where you have a conversation, forcing that conversation seems you're actually detrimental to the discovery. And interestingly enough, they said that, you know, on the product detail page, however, where I was not necessarily forced into a conversation, but I could actually ask for additional information while I actually found a product in that that that I'm interested in was actually good. It was something that makes sense. And they they added a word that I really like, which they or or I would say an expression here that they call the common sense place. So they said that it made sense for Rufus to be in that area. And what exactly is that common sense place? Well, it's what we call the conversion path. So the fact that the chatbot itself has never been part of the conversion path. It has never been an area where you discover, where you can visually see product, where you where you can see different variation and pricing. It's not like, you know, the the checkout where you already have your product and then you see recommendation that are tied to the product you have in the cart. It's not the search box with a lot which allow you to discover quickly. It's not like the recommendations, not listing pages. So it's kind of outside of it. But then when it's put on the product detail page, then it started to make a lot of sense. So this brings me back to what can we do about the search box. You know, the search box has always been in the in the conversion path. Can we evolve it? Can is it what's the future of that search box if the chatbot is kind of, you know, on the side? It kinda lives parallel to the discovery element. Can we augment it with GenAI? Because the the thing that makes the search box good is its speed. It is a thing you know, latency pretty much kills conversion. It is this has been proven so many time. And the moment that you start adding one second to the search response, you're losing customers. So what can we do with the search box? So we've done a few tests of generative AI in the search box. So, you know, I'm using here a grocery. Why grocery? Because it is, first of all, something that you want, you always want it to be fast in groceries because you might have to buy, you know, fifty products, and you don't want to have to waste time between the search on each product. Also, in a typical grocery scenario, the search engine will drive the search, listings, recommendations. It contains multiple products, different attributes, easy to get it wrong. But it also contains most of the time, on most, you know, most supermarket and groceries that I've visited online, there's always a section around recipe, around or or even, prebaked food. So there's kind of related, rich elements, to to the standard product, so additional documents that can give context to the products themselves. And let's say we start from the recipe. This is kind of the safe area. Instead of starting from the products, let's start from the recipe. And here in that kind of, you know, recipe homepage in a way, I can add, you know, a GenAI prompt box. And then, you know, I'll input a search. I host a summer barbecue in a park, and I need to cook for six people. What are your recommendations? You know, AI, please tell me, you know, what's what's going on. Here, you can have a pretty grounded, again, using, Caveo relevance generative answering, any source of rag. You will have a grounded answer based on the content that you have inside of your grocery store. You can even factor in inventory and local, mostly local availability from your store. So burger, grilled chicken skewers, hot dog and sausages, some sides, and whatnot. And then automatically, again, because you're within a search experience, we'll start showing you, you know, recommended recipes, as well as the facets on the on the left on the right. But then what if we add in one additional step to that Rag experience that we've talked about? So we add a third stage retrieval where we use semantic search to connect properly the response all the way to the products. So I'm searching for the same thing here. I host a summer barbecue in a park. I need to cook for six people, blah blah blah. And what do I receive? Well, the product or, in this case, actually, the categories related to what you're offering. So I'm offering classic burger. Well, you're probably interested into the burger aisle and category. I'm looking for a corn on the cob. There you go. Potato salads, ready meals inside. You know? Grilled veggie, fruits, and vegetable. So this idea of using semantic search to retrieve, in this case, products and categories is a great way to take an area of the site, the recipe, which was mostly used for SEO reason. Right? You want people to land on your site using these rich content and the recipe, which usually fare well on Google, and then you want them to continue their shopping once they read. So why don't you do the same thing here where, you know, when I search for a recipe, I will bring you slowly but surely into these different categories so you can start browsing. Why category more than products? Because the answer is quite is quite large. You know, I'm still at the point where I'm asking for a general question, and we have been used we've been led in the past, you know, that if I if I type a query, I should get result fast, which is kinda normal. Again, latency is king. But what if I ask a query here? Let me go back to my original query. What if I ask this query here, which is extremely broad, which, you know, does not necessarily yet indicate a clear product intention. I'm just looking for recommendation for a barbecue. It could go in in, you know, always. There's no way that you can scope the discovery enough to show specific product. Instead, showing categories is much safer. And then from there, you know, because you're not yet entirely sure if this shopper here is ready for product discovery and they might still be in that kind of education phase, you can still showcase, you know, these related recipes. Again, using semantic search, you're able to tie the keywords of the generated response with the content that you have, the rich content that you have. It does require a search engine that can support a unified index, which not all search engine will do, so adding structured content such as product and unstructured content such as recipe. But once you have, you know, a search engine that can do both, you can have that unified response to a generative answering, experience that has just happened before. But this leads me to our vision for the future of search, which is what we call the Caveo intent box. So it's a box that understands the intent and responds to it. It is not just a generative AI kind of based solution. It is based actually on a lot of AI techniques. In this case, lexical fuzzy matching, so being able to understand, again, the the the keyword, how they relate to the products, semantics understanding for similarity to increase the scope of the search, named entity recognition. This one is a more advanced technique, which is mostly being able to understand the different attributes on your product, how they relate to each other. So when I search, for example, for a vegetarian a vegetarian burger patty, the vegetarian element, is an attribute. It's actually a filtering attribute of the patty itself, which is the main product. So being able to understand what is what, what is the attribute, for example, what is a color, what is a size attribute, what is the main product I'm looking for, this allows us to understand, again, here the categories versus the product and what type of specific category or even other attribute that could be filtered that could be returned as the main category. So instead of returning, for example, all of the burger categories, I could say return all vegetarian products in all categories. So this is name entity recognition, also called NER. Intent detection using vector cosine similarity, quite a mouthful. It is understanding, and this is one of the most important element of this entire thing, understanding intent under the query, based on, you know, past patterns or even, based on the context of the sentence that this person has written or the prompt, as we call it. And then retrieval augmented generation, which I've explained several times, RAG, which is used, you know, to ground the GenAI response. So with all of these tools now in our toolbox, I can type a query in my Intentbox. So let's start with precision. I'm looking for a barbecue sauce. Semantic search, lexical matching, intent detection will return mostly barbecue sauce. Nothing so special there. I get fast, personalized product search with filters and sort. Typical answer. That being said, one thing is being added. It is query refinement. So this is using generative AI. It's using a natural language response that says, based on this query that has been launched, show the product first because we don't wanna waste time, and then take your time to come back to the user by proposing refinement. So would this product be better for chicken or fish? What would be the alternative with less sugar? Mostly, here, I'm proposing a conversation without forcing one. Now on the other side, let's start with a longer query. I host a summer barbecue in the park. I need to cook for six people. What are your recommendation? So barbecue, important element. It's in the park, so it's outside. It needs to be easy to to cook. Six people. What are the recommendations? Same answer that I got before. So I want, you know, burger, chicken skewers. I get the recipe on the left side, so rich content because you're still in that kind of education phase. I get, you know, the suggestion, the suggestion there at the bottom to refine my query. So here, again, you're into an education phase. We detected that you're not yet ready to convert. So we can take our time. We we can take our time to give, you know, a a a generated answer that will take a bit more time to render but will be much richer, will help you, will guide you towards your your kind of education journey. And then we start proposing product category. So what we want to do is to be as relevant as possible, understand it's the same search box, You're the same user, but we've detected with your query that you are now you have an intent of education. So we want to bring you as fast as possible yet as as relevant and guided as possible to a discovery experience where we will start to showcase these related category using named entity recognition and semantic search. So we will tie GenAI answers to product discovery. So what this leads us, you know, in as as a conclusion to this entire thing is the fact that we can definitely evolve the search box and integrate the new generative answering, the new guidance, the new relevance inside of that search box without falling into the trap of going for a robot like experience that is the good old chatbot or shopping assistant. So, you know, with the Coveo Intent Box, what we expect is that you will move from conversational to conversion. So it's no longer you know, the value of all of this technology is no longer how big of a conversation you can have with your shopper, but it's mostly how does it convert. Does it convert more or less? You know, what does it bring to the table? It moves from chatbot to discovery. So instead of being stuck in a chat that is not necessarily, you know, easy to use on mobile that takes, you know, kind of strange real estate on the site that requires, you know, a lot of back and forth, we're bringing you to product as quickly as we can. We're making sure that you discover what is the most valuable for you, which is the product. And we move from shopping assistant to an AI driven buying journey where AI is everywhere. It is on the main search. You know, if you if you type a precise, simple query, it leads you to the product that you need using the latest and greatest in AI and vector search and semantic search. But at the same time, if you look for, you know, education, you have a longer query which shows that, you know, you need some form of education, it brings you to an AI driven sorry, a generative experience, which is also part of that AI driven buying journey. So this is really where we see the future of the search box. It is not in the chat. It is not a shopping assistant. There will still need to be investment into your own discovery experience on your brand site, because Perplexity, ChatGPT, Google, Microsoft, or even Amazon will not replace this entire thing tomorrow. Maybe in the future. We never know what's gonna happen in the future. But tomorrow, it's yet another channel just like social media that will bring more people to your site. And when they reach your site, you need to convert. You need to bring them to the product as soon and as efficiently as possible, and AI is the way to do it. And on this, thank you very much for attending this webinar. I hope it was, insightful. I hope you learned, and it cleared a bit of confusion around GenAI and the future even of discovery of of your own brand, of your own solution, of your own investment. And if you want to pursue the conversation, my name is Simon Longstrain. I'm the VP of product at Caveo. You have my LinkedIn here on the site, and please do not hesitate to send me a direct message or to post and ask me to comment. It will be a pleasure to do so. Thank you very much, and have a great day.